2 * VXLAN: Virtual eXtensible Local Area Network
4 * Copyright (c) 2012-2013 Vyatta Inc.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/kernel.h>
14 #include <linux/types.h>
15 #include <linux/module.h>
16 #include <linux/errno.h>
17 #include <linux/slab.h>
18 #include <linux/skbuff.h>
19 #include <linux/rculist.h>
20 #include <linux/netdevice.h>
23 #include <linux/udp.h>
24 #include <linux/igmp.h>
25 #include <linux/etherdevice.h>
26 #include <linux/if_ether.h>
27 #include <linux/if_vlan.h>
28 #include <linux/hash.h>
29 #include <linux/ethtool.h>
31 #include <net/ndisc.h>
33 #include <net/ip_tunnels.h>
36 #include <net/udp_tunnel.h>
37 #include <net/rtnetlink.h>
38 #include <net/route.h>
39 #include <net/dsfield.h>
40 #include <net/inet_ecn.h>
41 #include <net/net_namespace.h>
42 #include <net/netns/generic.h>
43 #include <net/vxlan.h>
44 #include <net/protocol.h>
45 #include <net/udp_tunnel.h>
46 #if IS_ENABLED(CONFIG_IPV6)
48 #include <net/addrconf.h>
49 #include <net/ip6_tunnel.h>
50 #include <net/ip6_checksum.h>
53 #define VXLAN_VERSION "0.1"
55 #define PORT_HASH_BITS 8
56 #define PORT_HASH_SIZE (1<<PORT_HASH_BITS)
57 #define VNI_HASH_BITS 10
58 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
59 #define FDB_HASH_BITS 8
60 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
61 #define FDB_AGE_DEFAULT 300 /* 5 min */
62 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */
64 #define VXLAN_N_VID (1u << 24)
65 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
66 #define VXLAN_HLEN (sizeof(struct udphdr) + sizeof(struct vxlanhdr))
68 #define VXLAN_FLAGS 0x08000000 /* struct vxlanhdr.vx_flags required value. */
70 /* UDP port for VXLAN traffic.
71 * The IANA assigned port is 4789, but the Linux default is 8472
72 * for compatibility with early adopters.
74 static unsigned short vxlan_port __read_mostly = 8472;
75 module_param_named(udp_port, vxlan_port, ushort, 0444);
76 MODULE_PARM_DESC(udp_port, "Destination UDP port");
78 static bool log_ecn_error = true;
79 module_param(log_ecn_error, bool, 0644);
80 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
82 static int vxlan_net_id;
84 static const u8 all_zeros_mac[ETH_ALEN];
86 /* per-network namespace private data for this module */
88 struct list_head vxlan_list;
89 struct hlist_head sock_list[PORT_HASH_SIZE];
94 struct sockaddr_in sin;
95 struct sockaddr_in6 sin6;
100 union vxlan_addr remote_ip;
104 struct list_head list;
108 /* Forwarding table entry */
110 struct hlist_node hlist; /* linked list of entries */
112 unsigned long updated; /* jiffies */
114 struct list_head remotes;
115 u16 state; /* see ndm_state */
116 u8 flags; /* see ndm_flags */
117 u8 eth_addr[ETH_ALEN];
120 /* Pseudo network device */
122 struct hlist_node hlist; /* vni hash table */
123 struct list_head next; /* vxlan's per namespace list */
124 struct vxlan_sock *vn_sock; /* listening socket */
125 struct net_device *dev;
126 struct net *net; /* netns for packet i/o */
127 struct vxlan_rdst default_dst; /* default destination */
128 union vxlan_addr saddr; /* source address */
130 __u16 port_min; /* source port range */
132 __u8 tos; /* TOS override */
134 u32 flags; /* VXLAN_F_* in vxlan.h */
136 struct work_struct sock_work;
137 struct work_struct igmp_join;
138 struct work_struct igmp_leave;
140 unsigned long age_interval;
141 struct timer_list age_timer;
142 spinlock_t hash_lock;
143 unsigned int addrcnt;
144 unsigned int addrmax;
146 struct hlist_head fdb_head[FDB_HASH_SIZE];
149 /* salt for hash table */
150 static u32 vxlan_salt __read_mostly;
151 static struct workqueue_struct *vxlan_wq;
153 static void vxlan_sock_work(struct work_struct *work);
155 #if IS_ENABLED(CONFIG_IPV6)
157 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
159 if (a->sa.sa_family != b->sa.sa_family)
161 if (a->sa.sa_family == AF_INET6)
162 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
164 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
167 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
169 if (ipa->sa.sa_family == AF_INET6)
170 return ipv6_addr_any(&ipa->sin6.sin6_addr);
172 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
175 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
177 if (ipa->sa.sa_family == AF_INET6)
178 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
180 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
183 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
185 if (nla_len(nla) >= sizeof(struct in6_addr)) {
186 nla_memcpy(&ip->sin6.sin6_addr, nla, sizeof(struct in6_addr));
187 ip->sa.sa_family = AF_INET6;
189 } else if (nla_len(nla) >= sizeof(__be32)) {
190 ip->sin.sin_addr.s_addr = nla_get_be32(nla);
191 ip->sa.sa_family = AF_INET;
194 return -EAFNOSUPPORT;
198 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
199 const union vxlan_addr *ip)
201 if (ip->sa.sa_family == AF_INET6)
202 return nla_put(skb, attr, sizeof(struct in6_addr), &ip->sin6.sin6_addr);
204 return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
207 #else /* !CONFIG_IPV6 */
210 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
212 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
215 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
217 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
220 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
222 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
225 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
227 if (nla_len(nla) >= sizeof(struct in6_addr)) {
228 return -EAFNOSUPPORT;
229 } else if (nla_len(nla) >= sizeof(__be32)) {
230 ip->sin.sin_addr.s_addr = nla_get_be32(nla);
231 ip->sa.sa_family = AF_INET;
234 return -EAFNOSUPPORT;
238 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
239 const union vxlan_addr *ip)
241 return nla_put_be32(skb, attr, ip->sin.sin_addr.s_addr);
245 /* Virtual Network hash table head */
246 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, u32 id)
248 return &vs->vni_list[hash_32(id, VNI_HASH_BITS)];
251 /* Socket hash table head */
252 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
254 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
256 return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
259 /* First remote destination for a forwarding entry.
260 * Guaranteed to be non-NULL because remotes are never deleted.
262 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
264 return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
267 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
269 return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
272 /* Find VXLAN socket based on network namespace, address family and UDP port */
273 static struct vxlan_sock *vxlan_find_sock(struct net *net,
274 sa_family_t family, __be16 port)
276 struct vxlan_sock *vs;
278 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
279 if (inet_sk(vs->sock->sk)->inet_sport == port &&
280 inet_sk(vs->sock->sk)->sk.sk_family == family)
286 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, u32 id)
288 struct vxlan_dev *vxlan;
290 hlist_for_each_entry_rcu(vxlan, vni_head(vs, id), hlist) {
291 if (vxlan->default_dst.remote_vni == id)
298 /* Look up VNI in a per net namespace table */
299 static struct vxlan_dev *vxlan_find_vni(struct net *net, u32 id,
300 sa_family_t family, __be16 port)
302 struct vxlan_sock *vs;
304 vs = vxlan_find_sock(net, family, port);
308 return vxlan_vs_find_vni(vs, id);
311 /* Fill in neighbour message in skbuff. */
312 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
313 const struct vxlan_fdb *fdb,
314 u32 portid, u32 seq, int type, unsigned int flags,
315 const struct vxlan_rdst *rdst)
317 unsigned long now = jiffies;
318 struct nda_cacheinfo ci;
319 struct nlmsghdr *nlh;
321 bool send_ip, send_eth;
323 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
327 ndm = nlmsg_data(nlh);
328 memset(ndm, 0, sizeof(*ndm));
330 send_eth = send_ip = true;
332 if (type == RTM_GETNEIGH) {
333 ndm->ndm_family = AF_INET;
334 send_ip = !vxlan_addr_any(&rdst->remote_ip);
335 send_eth = !is_zero_ether_addr(fdb->eth_addr);
337 ndm->ndm_family = AF_BRIDGE;
338 ndm->ndm_state = fdb->state;
339 ndm->ndm_ifindex = vxlan->dev->ifindex;
340 ndm->ndm_flags = fdb->flags;
341 ndm->ndm_type = RTN_UNICAST;
343 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
344 goto nla_put_failure;
346 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
347 goto nla_put_failure;
349 if (rdst->remote_port && rdst->remote_port != vxlan->dst_port &&
350 nla_put_be16(skb, NDA_PORT, rdst->remote_port))
351 goto nla_put_failure;
352 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
353 nla_put_u32(skb, NDA_VNI, rdst->remote_vni))
354 goto nla_put_failure;
355 if (rdst->remote_ifindex &&
356 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
357 goto nla_put_failure;
359 ci.ndm_used = jiffies_to_clock_t(now - fdb->used);
360 ci.ndm_confirmed = 0;
361 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated);
364 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
365 goto nla_put_failure;
367 return nlmsg_end(skb, nlh);
370 nlmsg_cancel(skb, nlh);
374 static inline size_t vxlan_nlmsg_size(void)
376 return NLMSG_ALIGN(sizeof(struct ndmsg))
377 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
378 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
379 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
380 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
381 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
382 + nla_total_size(sizeof(struct nda_cacheinfo));
385 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
386 struct vxlan_rdst *rd, int type)
388 struct net *net = dev_net(vxlan->dev);
392 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
396 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
398 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
399 WARN_ON(err == -EMSGSIZE);
404 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
408 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
411 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
413 struct vxlan_dev *vxlan = netdev_priv(dev);
414 struct vxlan_fdb f = {
417 struct vxlan_rdst remote = {
418 .remote_ip = *ipa, /* goes to NDA_DST */
419 .remote_vni = VXLAN_N_VID,
422 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
425 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
427 struct vxlan_fdb f = {
430 struct vxlan_rdst remote = { };
432 memcpy(f.eth_addr, eth_addr, ETH_ALEN);
434 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
437 /* Hash Ethernet address */
438 static u32 eth_hash(const unsigned char *addr)
440 u64 value = get_unaligned((u64 *)addr);
442 /* only want 6 bytes */
448 return hash_64(value, FDB_HASH_BITS);
451 /* Hash chain to use given mac address */
452 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
455 return &vxlan->fdb_head[eth_hash(mac)];
458 /* Look up Ethernet address in forwarding table */
459 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
462 struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
465 hlist_for_each_entry_rcu(f, head, hlist) {
466 if (ether_addr_equal(mac, f->eth_addr))
473 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
478 f = __vxlan_find_mac(vxlan, mac);
485 /* caller should hold vxlan->hash_lock */
486 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
487 union vxlan_addr *ip, __be16 port,
488 __u32 vni, __u32 ifindex)
490 struct vxlan_rdst *rd;
492 list_for_each_entry(rd, &f->remotes, list) {
493 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
494 rd->remote_port == port &&
495 rd->remote_vni == vni &&
496 rd->remote_ifindex == ifindex)
503 /* Replace destination of unicast mac */
504 static int vxlan_fdb_replace(struct vxlan_fdb *f,
505 union vxlan_addr *ip, __be16 port, __u32 vni, __u32 ifindex)
507 struct vxlan_rdst *rd;
509 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
513 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
517 rd->remote_port = port;
518 rd->remote_vni = vni;
519 rd->remote_ifindex = ifindex;
523 /* Add/update destinations for multicast */
524 static int vxlan_fdb_append(struct vxlan_fdb *f,
525 union vxlan_addr *ip, __be16 port, __u32 vni,
526 __u32 ifindex, struct vxlan_rdst **rdp)
528 struct vxlan_rdst *rd;
530 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
534 rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
538 rd->remote_port = port;
539 rd->remote_vni = vni;
540 rd->remote_ifindex = ifindex;
542 list_add_tail_rcu(&rd->list, &f->remotes);
548 static struct sk_buff **vxlan_gro_receive(struct sk_buff **head, struct sk_buff *skb)
550 struct sk_buff *p, **pp = NULL;
551 struct vxlanhdr *vh, *vh2;
552 struct ethhdr *eh, *eh2;
553 unsigned int hlen, off_vx, off_eth;
554 const struct packet_offload *ptype;
558 off_vx = skb_gro_offset(skb);
559 hlen = off_vx + sizeof(*vh);
560 vh = skb_gro_header_fast(skb, off_vx);
561 if (skb_gro_header_hard(skb, hlen)) {
562 vh = skb_gro_header_slow(skb, hlen, off_vx);
566 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
567 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
569 off_eth = skb_gro_offset(skb);
570 hlen = off_eth + sizeof(*eh);
571 eh = skb_gro_header_fast(skb, off_eth);
572 if (skb_gro_header_hard(skb, hlen)) {
573 eh = skb_gro_header_slow(skb, hlen, off_eth);
580 for (p = *head; p; p = p->next) {
581 if (!NAPI_GRO_CB(p)->same_flow)
584 vh2 = (struct vxlanhdr *)(p->data + off_vx);
585 eh2 = (struct ethhdr *)(p->data + off_eth);
586 if (vh->vx_vni != vh2->vx_vni || compare_ether_header(eh, eh2)) {
587 NAPI_GRO_CB(p)->same_flow = 0;
595 ptype = gro_find_receive_by_type(type);
601 skb_gro_pull(skb, sizeof(*eh)); /* pull inner eth header */
602 skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
603 pp = ptype->callbacks.gro_receive(head, skb);
608 NAPI_GRO_CB(skb)->flush |= flush;
613 static int vxlan_gro_complete(struct sk_buff *skb, int nhoff)
616 struct packet_offload *ptype;
618 int vxlan_len = sizeof(struct vxlanhdr) + sizeof(struct ethhdr);
621 udp_tunnel_gro_complete(skb, nhoff);
623 eh = (struct ethhdr *)(skb->data + nhoff + sizeof(struct vxlanhdr));
627 ptype = gro_find_complete_by_type(type);
629 err = ptype->callbacks.gro_complete(skb, nhoff + vxlan_len);
635 /* Notify netdevs that UDP port started listening */
636 static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
638 struct net_device *dev;
639 struct sock *sk = vs->sock->sk;
640 struct net *net = sock_net(sk);
641 sa_family_t sa_family = sk->sk_family;
642 __be16 port = inet_sk(sk)->inet_sport;
645 if (sa_family == AF_INET) {
646 err = udp_add_offload(&vs->udp_offloads);
648 pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
652 for_each_netdev_rcu(net, dev) {
653 if (dev->netdev_ops->ndo_add_vxlan_port)
654 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
660 /* Notify netdevs that UDP port is no more listening */
661 static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
663 struct net_device *dev;
664 struct sock *sk = vs->sock->sk;
665 struct net *net = sock_net(sk);
666 sa_family_t sa_family = sk->sk_family;
667 __be16 port = inet_sk(sk)->inet_sport;
670 for_each_netdev_rcu(net, dev) {
671 if (dev->netdev_ops->ndo_del_vxlan_port)
672 dev->netdev_ops->ndo_del_vxlan_port(dev, sa_family,
677 if (sa_family == AF_INET)
678 udp_del_offload(&vs->udp_offloads);
681 /* Add new entry to forwarding table -- assumes lock held */
682 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
683 const u8 *mac, union vxlan_addr *ip,
684 __u16 state, __u16 flags,
685 __be16 port, __u32 vni, __u32 ifindex,
688 struct vxlan_rdst *rd = NULL;
692 f = __vxlan_find_mac(vxlan, mac);
694 if (flags & NLM_F_EXCL) {
695 netdev_dbg(vxlan->dev,
696 "lost race to create %pM\n", mac);
699 if (f->state != state) {
701 f->updated = jiffies;
704 if (f->flags != ndm_flags) {
705 f->flags = ndm_flags;
706 f->updated = jiffies;
709 if ((flags & NLM_F_REPLACE)) {
710 /* Only change unicasts */
711 if (!(is_multicast_ether_addr(f->eth_addr) ||
712 is_zero_ether_addr(f->eth_addr))) {
713 int rc = vxlan_fdb_replace(f, ip, port, vni,
722 if ((flags & NLM_F_APPEND) &&
723 (is_multicast_ether_addr(f->eth_addr) ||
724 is_zero_ether_addr(f->eth_addr))) {
725 int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
733 if (!(flags & NLM_F_CREATE))
736 if (vxlan->addrmax && vxlan->addrcnt >= vxlan->addrmax)
739 /* Disallow replace to add a multicast entry */
740 if ((flags & NLM_F_REPLACE) &&
741 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
744 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
745 f = kmalloc(sizeof(*f), GFP_ATOMIC);
751 f->flags = ndm_flags;
752 f->updated = f->used = jiffies;
753 INIT_LIST_HEAD(&f->remotes);
754 memcpy(f->eth_addr, mac, ETH_ALEN);
756 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
759 hlist_add_head_rcu(&f->hlist,
760 vxlan_fdb_head(vxlan, mac));
765 rd = first_remote_rtnl(f);
766 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
772 static void vxlan_fdb_free(struct rcu_head *head)
774 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
775 struct vxlan_rdst *rd, *nd;
777 list_for_each_entry_safe(rd, nd, &f->remotes, list)
782 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
784 netdev_dbg(vxlan->dev,
785 "delete %pM\n", f->eth_addr);
788 vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
790 hlist_del_rcu(&f->hlist);
791 call_rcu(&f->rcu, vxlan_fdb_free);
794 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
795 union vxlan_addr *ip, __be16 *port, u32 *vni, u32 *ifindex)
797 struct net *net = dev_net(vxlan->dev);
801 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
805 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
806 if (remote->sa.sa_family == AF_INET) {
807 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
808 ip->sa.sa_family = AF_INET;
809 #if IS_ENABLED(CONFIG_IPV6)
811 ip->sin6.sin6_addr = in6addr_any;
812 ip->sa.sa_family = AF_INET6;
818 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
820 *port = nla_get_be16(tb[NDA_PORT]);
822 *port = vxlan->dst_port;
826 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
828 *vni = nla_get_u32(tb[NDA_VNI]);
830 *vni = vxlan->default_dst.remote_vni;
833 if (tb[NDA_IFINDEX]) {
834 struct net_device *tdev;
836 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
838 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
839 tdev = __dev_get_by_index(net, *ifindex);
841 return -EADDRNOTAVAIL;
849 /* Add static entry (via netlink) */
850 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
851 struct net_device *dev,
852 const unsigned char *addr, u16 vid, u16 flags)
854 struct vxlan_dev *vxlan = netdev_priv(dev);
855 /* struct net *net = dev_net(vxlan->dev); */
861 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
862 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
867 if (tb[NDA_DST] == NULL)
870 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
874 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
875 return -EAFNOSUPPORT;
877 spin_lock_bh(&vxlan->hash_lock);
878 err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
879 port, vni, ifindex, ndm->ndm_flags);
880 spin_unlock_bh(&vxlan->hash_lock);
885 /* Delete entry (via netlink) */
886 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
887 struct net_device *dev,
888 const unsigned char *addr, u16 vid)
890 struct vxlan_dev *vxlan = netdev_priv(dev);
892 struct vxlan_rdst *rd = NULL;
898 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
904 spin_lock_bh(&vxlan->hash_lock);
905 f = vxlan_find_mac(vxlan, addr);
909 if (!vxlan_addr_any(&ip)) {
910 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
917 /* remove a destination if it's not the only one on the list,
918 * otherwise destroy the fdb entry
920 if (rd && !list_is_singular(&f->remotes)) {
921 list_del_rcu(&rd->list);
922 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
927 vxlan_fdb_destroy(vxlan, f);
930 spin_unlock_bh(&vxlan->hash_lock);
935 /* Dump forwarding table */
936 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
937 struct net_device *dev,
938 struct net_device *filter_dev, int idx)
940 struct vxlan_dev *vxlan = netdev_priv(dev);
943 for (h = 0; h < FDB_HASH_SIZE; ++h) {
947 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
948 struct vxlan_rdst *rd;
950 if (idx < cb->args[0])
953 list_for_each_entry_rcu(rd, &f->remotes, list) {
954 err = vxlan_fdb_info(skb, vxlan, f,
955 NETLINK_CB(cb->skb).portid,
970 /* Watch incoming packets to learn mapping between Ethernet address
971 * and Tunnel endpoint.
972 * Return true if packet is bogus and should be droppped.
974 static bool vxlan_snoop(struct net_device *dev,
975 union vxlan_addr *src_ip, const u8 *src_mac)
977 struct vxlan_dev *vxlan = netdev_priv(dev);
980 f = vxlan_find_mac(vxlan, src_mac);
982 struct vxlan_rdst *rdst = first_remote_rcu(f);
984 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
987 /* Don't migrate static entries, drop packets */
988 if (f->state & NUD_NOARP)
993 "%pM migrated from %pIS to %pIS\n",
994 src_mac, &rdst->remote_ip, &src_ip);
996 rdst->remote_ip = *src_ip;
997 f->updated = jiffies;
998 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
1000 /* learned new entry */
1001 spin_lock(&vxlan->hash_lock);
1003 /* close off race between vxlan_flush and incoming packets */
1004 if (netif_running(dev))
1005 vxlan_fdb_create(vxlan, src_mac, src_ip,
1007 NLM_F_EXCL|NLM_F_CREATE,
1009 vxlan->default_dst.remote_vni,
1011 spin_unlock(&vxlan->hash_lock);
1017 /* See if multicast group is already in use by other ID */
1018 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1020 struct vxlan_dev *vxlan;
1022 /* The vxlan_sock is only used by dev, leaving group has
1023 * no effect on other vxlan devices.
1025 if (atomic_read(&dev->vn_sock->refcnt) == 1)
1028 list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1029 if (!netif_running(vxlan->dev) || vxlan == dev)
1032 if (vxlan->vn_sock != dev->vn_sock)
1035 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1036 &dev->default_dst.remote_ip))
1039 if (vxlan->default_dst.remote_ifindex !=
1040 dev->default_dst.remote_ifindex)
1049 static void vxlan_sock_hold(struct vxlan_sock *vs)
1051 atomic_inc(&vs->refcnt);
1054 void vxlan_sock_release(struct vxlan_sock *vs)
1056 struct sock *sk = vs->sock->sk;
1057 struct net *net = sock_net(sk);
1058 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
1060 if (!atomic_dec_and_test(&vs->refcnt))
1063 spin_lock(&vn->sock_lock);
1064 hlist_del_rcu(&vs->hlist);
1065 vxlan_notify_del_rx_port(vs);
1066 spin_unlock(&vn->sock_lock);
1068 queue_work(vxlan_wq, &vs->del_work);
1070 EXPORT_SYMBOL_GPL(vxlan_sock_release);
1072 /* Callback to update multicast group membership when first VNI on
1073 * multicast asddress is brought up
1074 * Done as workqueue because ip_mc_join_group acquires RTNL.
1076 static void vxlan_igmp_join(struct work_struct *work)
1078 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_join);
1079 struct vxlan_sock *vs = vxlan->vn_sock;
1080 struct sock *sk = vs->sock->sk;
1081 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1082 int ifindex = vxlan->default_dst.remote_ifindex;
1085 if (ip->sa.sa_family == AF_INET) {
1086 struct ip_mreqn mreq = {
1087 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1088 .imr_ifindex = ifindex,
1091 ip_mc_join_group(sk, &mreq);
1092 #if IS_ENABLED(CONFIG_IPV6)
1094 ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1095 &ip->sin6.sin6_addr);
1100 vxlan_sock_release(vs);
1101 dev_put(vxlan->dev);
1104 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1105 static void vxlan_igmp_leave(struct work_struct *work)
1107 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, igmp_leave);
1108 struct vxlan_sock *vs = vxlan->vn_sock;
1109 struct sock *sk = vs->sock->sk;
1110 union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1111 int ifindex = vxlan->default_dst.remote_ifindex;
1114 if (ip->sa.sa_family == AF_INET) {
1115 struct ip_mreqn mreq = {
1116 .imr_multiaddr.s_addr = ip->sin.sin_addr.s_addr,
1117 .imr_ifindex = ifindex,
1120 ip_mc_leave_group(sk, &mreq);
1121 #if IS_ENABLED(CONFIG_IPV6)
1123 ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1124 &ip->sin6.sin6_addr);
1130 vxlan_sock_release(vs);
1131 dev_put(vxlan->dev);
1134 /* Callback from net/ipv4/udp.c to receive packets */
1135 static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
1137 struct vxlan_sock *vs;
1138 struct vxlanhdr *vxh;
1140 /* Need Vxlan and inner Ethernet header to be present */
1141 if (!pskb_may_pull(skb, VXLAN_HLEN))
1144 /* Return packets with reserved bits set */
1145 vxh = (struct vxlanhdr *)(udp_hdr(skb) + 1);
1146 if (vxh->vx_flags != htonl(VXLAN_FLAGS) ||
1147 (vxh->vx_vni & htonl(0xff))) {
1148 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1149 ntohl(vxh->vx_flags), ntohl(vxh->vx_vni));
1153 if (iptunnel_pull_header(skb, VXLAN_HLEN, htons(ETH_P_TEB)))
1156 vs = rcu_dereference_sk_user_data(sk);
1160 vs->rcv(vs, skb, vxh->vx_vni);
1164 /* Consume bad packet */
1169 /* Return non vxlan pkt */
1173 static void vxlan_rcv(struct vxlan_sock *vs,
1174 struct sk_buff *skb, __be32 vx_vni)
1176 struct iphdr *oip = NULL;
1177 struct ipv6hdr *oip6 = NULL;
1178 struct vxlan_dev *vxlan;
1179 struct pcpu_sw_netstats *stats;
1180 union vxlan_addr saddr;
1183 union vxlan_addr *remote_ip;
1185 vni = ntohl(vx_vni) >> 8;
1186 /* Is this VNI defined? */
1187 vxlan = vxlan_vs_find_vni(vs, vni);
1191 remote_ip = &vxlan->default_dst.remote_ip;
1192 skb_reset_mac_header(skb);
1193 skb_scrub_packet(skb, !net_eq(vxlan->net, dev_net(vxlan->dev)));
1194 skb->protocol = eth_type_trans(skb, vxlan->dev);
1195 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1197 /* Ignore packet loops (and multicast echo) */
1198 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1201 /* Re-examine inner Ethernet packet */
1202 if (remote_ip->sa.sa_family == AF_INET) {
1204 saddr.sin.sin_addr.s_addr = oip->saddr;
1205 saddr.sa.sa_family = AF_INET;
1206 #if IS_ENABLED(CONFIG_IPV6)
1208 oip6 = ipv6_hdr(skb);
1209 saddr.sin6.sin6_addr = oip6->saddr;
1210 saddr.sa.sa_family = AF_INET6;
1214 if ((vxlan->flags & VXLAN_F_LEARN) &&
1215 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1218 skb_reset_network_header(skb);
1221 err = IP6_ECN_decapsulate(oip6, skb);
1223 err = IP_ECN_decapsulate(oip, skb);
1225 if (unlikely(err)) {
1226 if (log_ecn_error) {
1228 net_info_ratelimited("non-ECT from %pI6\n",
1231 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1232 &oip->saddr, oip->tos);
1235 ++vxlan->dev->stats.rx_frame_errors;
1236 ++vxlan->dev->stats.rx_errors;
1241 stats = this_cpu_ptr(vxlan->dev->tstats);
1242 u64_stats_update_begin(&stats->syncp);
1243 stats->rx_packets++;
1244 stats->rx_bytes += skb->len;
1245 u64_stats_update_end(&stats->syncp);
1251 /* Consume bad packet */
1255 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1257 struct vxlan_dev *vxlan = netdev_priv(dev);
1258 struct arphdr *parp;
1261 struct neighbour *n;
1263 if (dev->flags & IFF_NOARP)
1266 if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1267 dev->stats.tx_dropped++;
1270 parp = arp_hdr(skb);
1272 if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1273 parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1274 parp->ar_pro != htons(ETH_P_IP) ||
1275 parp->ar_op != htons(ARPOP_REQUEST) ||
1276 parp->ar_hln != dev->addr_len ||
1279 arpptr = (u8 *)parp + sizeof(struct arphdr);
1281 arpptr += dev->addr_len; /* sha */
1282 memcpy(&sip, arpptr, sizeof(sip));
1283 arpptr += sizeof(sip);
1284 arpptr += dev->addr_len; /* tha */
1285 memcpy(&tip, arpptr, sizeof(tip));
1287 if (ipv4_is_loopback(tip) ||
1288 ipv4_is_multicast(tip))
1291 n = neigh_lookup(&arp_tbl, &tip, dev);
1294 struct vxlan_fdb *f;
1295 struct sk_buff *reply;
1297 if (!(n->nud_state & NUD_CONNECTED)) {
1302 f = vxlan_find_mac(vxlan, n->ha);
1303 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1304 /* bridge-local neighbor */
1309 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1317 skb_reset_mac_header(reply);
1318 __skb_pull(reply, skb_network_offset(reply));
1319 reply->ip_summed = CHECKSUM_UNNECESSARY;
1320 reply->pkt_type = PACKET_HOST;
1322 if (netif_rx_ni(reply) == NET_RX_DROP)
1323 dev->stats.rx_dropped++;
1324 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1325 union vxlan_addr ipa = {
1326 .sin.sin_addr.s_addr = tip,
1327 .sin.sin_family = AF_INET,
1330 vxlan_ip_miss(dev, &ipa);
1334 return NETDEV_TX_OK;
1337 #if IS_ENABLED(CONFIG_IPV6)
1338 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1339 struct neighbour *n, bool isrouter)
1341 struct net_device *dev = request->dev;
1342 struct sk_buff *reply;
1343 struct nd_msg *ns, *na;
1344 struct ipv6hdr *pip6;
1346 int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1353 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1354 sizeof(*na) + na_olen + dev->needed_tailroom;
1355 reply = alloc_skb(len, GFP_ATOMIC);
1359 reply->protocol = htons(ETH_P_IPV6);
1361 skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1362 skb_push(reply, sizeof(struct ethhdr));
1363 skb_set_mac_header(reply, 0);
1365 ns = (struct nd_msg *)skb_transport_header(request);
1367 daddr = eth_hdr(request)->h_source;
1368 ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1369 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1370 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1371 daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1376 /* Ethernet header */
1377 ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1378 ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1379 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1380 reply->protocol = htons(ETH_P_IPV6);
1382 skb_pull(reply, sizeof(struct ethhdr));
1383 skb_set_network_header(reply, 0);
1384 skb_put(reply, sizeof(struct ipv6hdr));
1388 pip6 = ipv6_hdr(reply);
1389 memset(pip6, 0, sizeof(struct ipv6hdr));
1391 pip6->priority = ipv6_hdr(request)->priority;
1392 pip6->nexthdr = IPPROTO_ICMPV6;
1393 pip6->hop_limit = 255;
1394 pip6->daddr = ipv6_hdr(request)->saddr;
1395 pip6->saddr = *(struct in6_addr *)n->primary_key;
1397 skb_pull(reply, sizeof(struct ipv6hdr));
1398 skb_set_transport_header(reply, 0);
1400 na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1402 /* Neighbor Advertisement */
1403 memset(na, 0, sizeof(*na)+na_olen);
1404 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1405 na->icmph.icmp6_router = isrouter;
1406 na->icmph.icmp6_override = 1;
1407 na->icmph.icmp6_solicited = 1;
1408 na->target = ns->target;
1409 ether_addr_copy(&na->opt[2], n->ha);
1410 na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1411 na->opt[1] = na_olen >> 3;
1413 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1414 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1415 csum_partial(na, sizeof(*na)+na_olen, 0));
1417 pip6->payload_len = htons(sizeof(*na)+na_olen);
1419 skb_push(reply, sizeof(struct ipv6hdr));
1421 reply->ip_summed = CHECKSUM_UNNECESSARY;
1426 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1428 struct vxlan_dev *vxlan = netdev_priv(dev);
1430 const struct ipv6hdr *iphdr;
1431 const struct in6_addr *saddr, *daddr;
1432 struct neighbour *n;
1433 struct inet6_dev *in6_dev;
1435 in6_dev = __in6_dev_get(dev);
1439 iphdr = ipv6_hdr(skb);
1440 saddr = &iphdr->saddr;
1441 daddr = &iphdr->daddr;
1443 msg = (struct nd_msg *)skb_transport_header(skb);
1444 if (msg->icmph.icmp6_code != 0 ||
1445 msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1448 if (ipv6_addr_loopback(daddr) ||
1449 ipv6_addr_is_multicast(&msg->target))
1452 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1455 struct vxlan_fdb *f;
1456 struct sk_buff *reply;
1458 if (!(n->nud_state & NUD_CONNECTED)) {
1463 f = vxlan_find_mac(vxlan, n->ha);
1464 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1465 /* bridge-local neighbor */
1470 reply = vxlan_na_create(skb, n,
1471 !!(f ? f->flags & NTF_ROUTER : 0));
1478 if (netif_rx_ni(reply) == NET_RX_DROP)
1479 dev->stats.rx_dropped++;
1481 } else if (vxlan->flags & VXLAN_F_L3MISS) {
1482 union vxlan_addr ipa = {
1483 .sin6.sin6_addr = msg->target,
1484 .sin6.sin6_family = AF_INET6,
1487 vxlan_ip_miss(dev, &ipa);
1492 return NETDEV_TX_OK;
1496 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1498 struct vxlan_dev *vxlan = netdev_priv(dev);
1499 struct neighbour *n;
1501 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1505 switch (ntohs(eth_hdr(skb)->h_proto)) {
1510 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1513 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1514 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1515 union vxlan_addr ipa = {
1516 .sin.sin_addr.s_addr = pip->daddr,
1517 .sin.sin_family = AF_INET,
1520 vxlan_ip_miss(dev, &ipa);
1526 #if IS_ENABLED(CONFIG_IPV6)
1529 struct ipv6hdr *pip6;
1531 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1533 pip6 = ipv6_hdr(skb);
1534 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1535 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1536 union vxlan_addr ipa = {
1537 .sin6.sin6_addr = pip6->daddr,
1538 .sin6.sin6_family = AF_INET6,
1541 vxlan_ip_miss(dev, &ipa);
1555 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1557 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1559 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1568 #if IS_ENABLED(CONFIG_IPV6)
1569 static int vxlan6_xmit_skb(struct vxlan_sock *vs,
1570 struct dst_entry *dst, struct sk_buff *skb,
1571 struct net_device *dev, struct in6_addr *saddr,
1572 struct in6_addr *daddr, __u8 prio, __u8 ttl,
1573 __be16 src_port, __be16 dst_port, __be32 vni,
1576 struct vxlanhdr *vxh;
1579 bool udp_sum = !udp_get_no_check6_tx(vs->sock->sk);
1581 skb = udp_tunnel_handle_offloads(skb, udp_sum);
1587 skb_scrub_packet(skb, xnet);
1589 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1590 + VXLAN_HLEN + sizeof(struct ipv6hdr)
1591 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1593 /* Need space for new headers (invalidates iph ptr) */
1594 err = skb_cow_head(skb, min_headroom);
1595 if (unlikely(err)) {
1600 skb = vlan_hwaccel_push_inside(skb);
1601 if (WARN_ON(!skb)) {
1606 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1607 vxh->vx_flags = htonl(VXLAN_FLAGS);
1610 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1612 udp_tunnel6_xmit_skb(vs->sock, dst, skb, dev, saddr, daddr, prio,
1613 ttl, src_port, dst_port);
1621 int vxlan_xmit_skb(struct vxlan_sock *vs,
1622 struct rtable *rt, struct sk_buff *skb,
1623 __be32 src, __be32 dst, __u8 tos, __u8 ttl, __be16 df,
1624 __be16 src_port, __be16 dst_port, __be32 vni, bool xnet)
1626 struct vxlanhdr *vxh;
1629 bool udp_sum = !vs->sock->sk->sk_no_check_tx;
1631 skb = udp_tunnel_handle_offloads(skb, udp_sum);
1633 return PTR_ERR(skb);
1635 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1636 + VXLAN_HLEN + sizeof(struct iphdr)
1637 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1639 /* Need space for new headers (invalidates iph ptr) */
1640 err = skb_cow_head(skb, min_headroom);
1641 if (unlikely(err)) {
1646 skb = vlan_hwaccel_push_inside(skb);
1650 vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1651 vxh->vx_flags = htonl(VXLAN_FLAGS);
1654 skb_set_inner_protocol(skb, htons(ETH_P_TEB));
1656 return udp_tunnel_xmit_skb(vs->sock, rt, skb, src, dst, tos,
1657 ttl, df, src_port, dst_port, xnet);
1659 EXPORT_SYMBOL_GPL(vxlan_xmit_skb);
1661 /* Bypass encapsulation if the destination is local */
1662 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1663 struct vxlan_dev *dst_vxlan)
1665 struct pcpu_sw_netstats *tx_stats, *rx_stats;
1666 union vxlan_addr loopback;
1667 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1668 struct net_device *dev = skb->dev;
1671 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1672 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1673 skb->pkt_type = PACKET_HOST;
1674 skb->encapsulation = 0;
1675 skb->dev = dst_vxlan->dev;
1676 __skb_pull(skb, skb_network_offset(skb));
1678 if (remote_ip->sa.sa_family == AF_INET) {
1679 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1680 loopback.sa.sa_family = AF_INET;
1681 #if IS_ENABLED(CONFIG_IPV6)
1683 loopback.sin6.sin6_addr = in6addr_loopback;
1684 loopback.sa.sa_family = AF_INET6;
1688 if (dst_vxlan->flags & VXLAN_F_LEARN)
1689 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1691 u64_stats_update_begin(&tx_stats->syncp);
1692 tx_stats->tx_packets++;
1693 tx_stats->tx_bytes += len;
1694 u64_stats_update_end(&tx_stats->syncp);
1696 if (netif_rx(skb) == NET_RX_SUCCESS) {
1697 u64_stats_update_begin(&rx_stats->syncp);
1698 rx_stats->rx_packets++;
1699 rx_stats->rx_bytes += len;
1700 u64_stats_update_end(&rx_stats->syncp);
1702 dev->stats.rx_dropped++;
1706 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1707 struct vxlan_rdst *rdst, bool did_rsc)
1709 struct vxlan_dev *vxlan = netdev_priv(dev);
1710 struct rtable *rt = NULL;
1711 const struct iphdr *old_iph;
1713 union vxlan_addr *dst;
1714 __be16 src_port = 0, dst_port;
1720 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->dst_port;
1721 vni = rdst->remote_vni;
1722 dst = &rdst->remote_ip;
1724 if (vxlan_addr_any(dst)) {
1726 /* short-circuited back to local bridge */
1727 vxlan_encap_bypass(skb, vxlan, vxlan);
1733 old_iph = ip_hdr(skb);
1736 if (!ttl && vxlan_addr_multicast(dst))
1741 tos = ip_tunnel_get_dsfield(old_iph, skb);
1743 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->port_min,
1744 vxlan->port_max, true);
1746 if (dst->sa.sa_family == AF_INET) {
1747 memset(&fl4, 0, sizeof(fl4));
1748 fl4.flowi4_oif = rdst->remote_ifindex;
1749 fl4.flowi4_tos = RT_TOS(tos);
1750 fl4.daddr = dst->sin.sin_addr.s_addr;
1751 fl4.saddr = vxlan->saddr.sin.sin_addr.s_addr;
1753 rt = ip_route_output_key(vxlan->net, &fl4);
1755 netdev_dbg(dev, "no route to %pI4\n",
1756 &dst->sin.sin_addr.s_addr);
1757 dev->stats.tx_carrier_errors++;
1761 if (rt->dst.dev == dev) {
1762 netdev_dbg(dev, "circular route to %pI4\n",
1763 &dst->sin.sin_addr.s_addr);
1764 dev->stats.collisions++;
1768 /* Bypass encapsulation if the destination is local */
1769 if (rt->rt_flags & RTCF_LOCAL &&
1770 !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1771 struct vxlan_dev *dst_vxlan;
1774 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1775 dst->sa.sa_family, dst_port);
1778 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1782 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
1783 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
1785 err = vxlan_xmit_skb(vxlan->vn_sock, rt, skb,
1786 fl4.saddr, dst->sin.sin_addr.s_addr,
1787 tos, ttl, df, src_port, dst_port,
1789 !net_eq(vxlan->net, dev_net(vxlan->dev)));
1791 /* skb is already freed. */
1796 iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
1797 #if IS_ENABLED(CONFIG_IPV6)
1799 struct sock *sk = vxlan->vn_sock->sock->sk;
1800 struct dst_entry *ndst;
1804 memset(&fl6, 0, sizeof(fl6));
1805 fl6.flowi6_oif = rdst->remote_ifindex;
1806 fl6.daddr = dst->sin6.sin6_addr;
1807 fl6.saddr = vxlan->saddr.sin6.sin6_addr;
1808 fl6.flowi6_proto = IPPROTO_UDP;
1810 if (ipv6_stub->ipv6_dst_lookup(sk, &ndst, &fl6)) {
1811 netdev_dbg(dev, "no route to %pI6\n",
1812 &dst->sin6.sin6_addr);
1813 dev->stats.tx_carrier_errors++;
1817 if (ndst->dev == dev) {
1818 netdev_dbg(dev, "circular route to %pI6\n",
1819 &dst->sin6.sin6_addr);
1821 dev->stats.collisions++;
1825 /* Bypass encapsulation if the destination is local */
1826 flags = ((struct rt6_info *)ndst)->rt6i_flags;
1827 if (flags & RTF_LOCAL &&
1828 !(flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1829 struct vxlan_dev *dst_vxlan;
1832 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1833 dst->sa.sa_family, dst_port);
1836 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1840 ttl = ttl ? : ip6_dst_hoplimit(ndst);
1842 err = vxlan6_xmit_skb(vxlan->vn_sock, ndst, skb,
1843 dev, &fl6.saddr, &fl6.daddr, 0, ttl,
1844 src_port, dst_port, htonl(vni << 8),
1845 !net_eq(vxlan->net, dev_net(vxlan->dev)));
1852 dev->stats.tx_dropped++;
1858 dev->stats.tx_errors++;
1863 /* Transmit local packets over Vxlan
1865 * Outer IP header inherits ECN and DF from inner header.
1866 * Outer UDP destination is the VXLAN assigned port.
1867 * source port is based on hash of flow
1869 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
1871 struct vxlan_dev *vxlan = netdev_priv(dev);
1873 bool did_rsc = false;
1874 struct vxlan_rdst *rdst, *fdst = NULL;
1875 struct vxlan_fdb *f;
1877 skb_reset_mac_header(skb);
1880 if ((vxlan->flags & VXLAN_F_PROXY)) {
1881 if (ntohs(eth->h_proto) == ETH_P_ARP)
1882 return arp_reduce(dev, skb);
1883 #if IS_ENABLED(CONFIG_IPV6)
1884 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
1885 pskb_may_pull(skb, sizeof(struct ipv6hdr)
1886 + sizeof(struct nd_msg)) &&
1887 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
1890 msg = (struct nd_msg *)skb_transport_header(skb);
1891 if (msg->icmph.icmp6_code == 0 &&
1892 msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
1893 return neigh_reduce(dev, skb);
1899 f = vxlan_find_mac(vxlan, eth->h_dest);
1902 if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
1903 (ntohs(eth->h_proto) == ETH_P_IP ||
1904 ntohs(eth->h_proto) == ETH_P_IPV6)) {
1905 did_rsc = route_shortcircuit(dev, skb);
1907 f = vxlan_find_mac(vxlan, eth->h_dest);
1911 f = vxlan_find_mac(vxlan, all_zeros_mac);
1913 if ((vxlan->flags & VXLAN_F_L2MISS) &&
1914 !is_multicast_ether_addr(eth->h_dest))
1915 vxlan_fdb_miss(vxlan, eth->h_dest);
1917 dev->stats.tx_dropped++;
1919 return NETDEV_TX_OK;
1923 list_for_each_entry_rcu(rdst, &f->remotes, list) {
1924 struct sk_buff *skb1;
1930 skb1 = skb_clone(skb, GFP_ATOMIC);
1932 vxlan_xmit_one(skb1, dev, rdst, did_rsc);
1936 vxlan_xmit_one(skb, dev, fdst, did_rsc);
1939 return NETDEV_TX_OK;
1942 /* Walk the forwarding table and purge stale entries */
1943 static void vxlan_cleanup(unsigned long arg)
1945 struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
1946 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
1949 if (!netif_running(vxlan->dev))
1952 spin_lock_bh(&vxlan->hash_lock);
1953 for (h = 0; h < FDB_HASH_SIZE; ++h) {
1954 struct hlist_node *p, *n;
1955 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
1957 = container_of(p, struct vxlan_fdb, hlist);
1958 unsigned long timeout;
1960 if (f->state & NUD_PERMANENT)
1963 timeout = f->used + vxlan->age_interval * HZ;
1964 if (time_before_eq(timeout, jiffies)) {
1965 netdev_dbg(vxlan->dev,
1966 "garbage collect %pM\n",
1968 f->state = NUD_STALE;
1969 vxlan_fdb_destroy(vxlan, f);
1970 } else if (time_before(timeout, next_timer))
1971 next_timer = timeout;
1974 spin_unlock_bh(&vxlan->hash_lock);
1976 mod_timer(&vxlan->age_timer, next_timer);
1979 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
1981 __u32 vni = vxlan->default_dst.remote_vni;
1983 vxlan->vn_sock = vs;
1984 hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
1987 /* Setup stats when device is created */
1988 static int vxlan_init(struct net_device *dev)
1990 struct vxlan_dev *vxlan = netdev_priv(dev);
1991 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
1992 struct vxlan_sock *vs;
1993 bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
1995 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1999 spin_lock(&vn->sock_lock);
2000 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2002 if (vs && atomic_add_unless(&vs->refcnt, 1, 0)) {
2003 /* If we have a socket with same port already, reuse it */
2004 vxlan_vs_add_dev(vs, vxlan);
2006 /* otherwise make new socket outside of RTNL */
2008 queue_work(vxlan_wq, &vxlan->sock_work);
2010 spin_unlock(&vn->sock_lock);
2015 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2017 struct vxlan_fdb *f;
2019 spin_lock_bh(&vxlan->hash_lock);
2020 f = __vxlan_find_mac(vxlan, all_zeros_mac);
2022 vxlan_fdb_destroy(vxlan, f);
2023 spin_unlock_bh(&vxlan->hash_lock);
2026 static void vxlan_uninit(struct net_device *dev)
2028 struct vxlan_dev *vxlan = netdev_priv(dev);
2029 struct vxlan_sock *vs = vxlan->vn_sock;
2031 vxlan_fdb_delete_default(vxlan);
2034 vxlan_sock_release(vs);
2035 free_percpu(dev->tstats);
2038 /* Start ageing timer and join group when device is brought up */
2039 static int vxlan_open(struct net_device *dev)
2041 struct vxlan_dev *vxlan = netdev_priv(dev);
2042 struct vxlan_sock *vs = vxlan->vn_sock;
2044 /* socket hasn't been created */
2048 if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2049 vxlan_sock_hold(vs);
2051 queue_work(vxlan_wq, &vxlan->igmp_join);
2054 if (vxlan->age_interval)
2055 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2060 /* Purge the forwarding table */
2061 static void vxlan_flush(struct vxlan_dev *vxlan)
2065 spin_lock_bh(&vxlan->hash_lock);
2066 for (h = 0; h < FDB_HASH_SIZE; ++h) {
2067 struct hlist_node *p, *n;
2068 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2070 = container_of(p, struct vxlan_fdb, hlist);
2071 /* the all_zeros_mac entry is deleted at vxlan_uninit */
2072 if (!is_zero_ether_addr(f->eth_addr))
2073 vxlan_fdb_destroy(vxlan, f);
2076 spin_unlock_bh(&vxlan->hash_lock);
2079 /* Cleanup timer and forwarding table on shutdown */
2080 static int vxlan_stop(struct net_device *dev)
2082 struct vxlan_dev *vxlan = netdev_priv(dev);
2083 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2084 struct vxlan_sock *vs = vxlan->vn_sock;
2086 if (vs && vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2087 !vxlan_group_used(vn, vxlan)) {
2088 vxlan_sock_hold(vs);
2090 queue_work(vxlan_wq, &vxlan->igmp_leave);
2093 del_timer_sync(&vxlan->age_timer);
2100 /* Stub, nothing needs to be done. */
2101 static void vxlan_set_multicast_list(struct net_device *dev)
2105 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2107 struct vxlan_dev *vxlan = netdev_priv(dev);
2108 struct vxlan_rdst *dst = &vxlan->default_dst;
2109 struct net_device *lowerdev;
2112 lowerdev = __dev_get_by_index(vxlan->net, dst->remote_ifindex);
2113 if (lowerdev == NULL)
2114 return eth_change_mtu(dev, new_mtu);
2116 if (dst->remote_ip.sa.sa_family == AF_INET6)
2117 max_mtu = lowerdev->mtu - VXLAN6_HEADROOM;
2119 max_mtu = lowerdev->mtu - VXLAN_HEADROOM;
2121 if (new_mtu < 68 || new_mtu > max_mtu)
2128 static const struct net_device_ops vxlan_netdev_ops = {
2129 .ndo_init = vxlan_init,
2130 .ndo_uninit = vxlan_uninit,
2131 .ndo_open = vxlan_open,
2132 .ndo_stop = vxlan_stop,
2133 .ndo_start_xmit = vxlan_xmit,
2134 .ndo_get_stats64 = ip_tunnel_get_stats64,
2135 .ndo_set_rx_mode = vxlan_set_multicast_list,
2136 .ndo_change_mtu = vxlan_change_mtu,
2137 .ndo_validate_addr = eth_validate_addr,
2138 .ndo_set_mac_address = eth_mac_addr,
2139 .ndo_fdb_add = vxlan_fdb_add,
2140 .ndo_fdb_del = vxlan_fdb_delete,
2141 .ndo_fdb_dump = vxlan_fdb_dump,
2144 /* Info for udev, that this is a virtual tunnel endpoint */
2145 static struct device_type vxlan_type = {
2149 /* Calls the ndo_add_vxlan_port of the caller in order to
2150 * supply the listening VXLAN udp ports. Callers are expected
2151 * to implement the ndo_add_vxlan_port.
2153 void vxlan_get_rx_port(struct net_device *dev)
2155 struct vxlan_sock *vs;
2156 struct net *net = dev_net(dev);
2157 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2158 sa_family_t sa_family;
2162 spin_lock(&vn->sock_lock);
2163 for (i = 0; i < PORT_HASH_SIZE; ++i) {
2164 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2165 port = inet_sk(vs->sock->sk)->inet_sport;
2166 sa_family = vs->sock->sk->sk_family;
2167 dev->netdev_ops->ndo_add_vxlan_port(dev, sa_family,
2171 spin_unlock(&vn->sock_lock);
2173 EXPORT_SYMBOL_GPL(vxlan_get_rx_port);
2175 /* Initialize the device structure. */
2176 static void vxlan_setup(struct net_device *dev)
2178 struct vxlan_dev *vxlan = netdev_priv(dev);
2181 eth_hw_addr_random(dev);
2183 if (vxlan->default_dst.remote_ip.sa.sa_family == AF_INET6)
2184 dev->needed_headroom = ETH_HLEN + VXLAN6_HEADROOM;
2186 dev->needed_headroom = ETH_HLEN + VXLAN_HEADROOM;
2188 dev->netdev_ops = &vxlan_netdev_ops;
2189 dev->destructor = free_netdev;
2190 SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2192 dev->tx_queue_len = 0;
2193 dev->features |= NETIF_F_LLTX;
2194 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2195 dev->features |= NETIF_F_RXCSUM;
2196 dev->features |= NETIF_F_GSO_SOFTWARE;
2198 dev->vlan_features = dev->features;
2199 dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2200 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2201 dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2202 dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2203 netif_keep_dst(dev);
2204 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2206 INIT_LIST_HEAD(&vxlan->next);
2207 spin_lock_init(&vxlan->hash_lock);
2208 INIT_WORK(&vxlan->igmp_join, vxlan_igmp_join);
2209 INIT_WORK(&vxlan->igmp_leave, vxlan_igmp_leave);
2210 INIT_WORK(&vxlan->sock_work, vxlan_sock_work);
2212 init_timer_deferrable(&vxlan->age_timer);
2213 vxlan->age_timer.function = vxlan_cleanup;
2214 vxlan->age_timer.data = (unsigned long) vxlan;
2216 vxlan->dst_port = htons(vxlan_port);
2220 for (h = 0; h < FDB_HASH_SIZE; ++h)
2221 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2224 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2225 [IFLA_VXLAN_ID] = { .type = NLA_U32 },
2226 [IFLA_VXLAN_GROUP] = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2227 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) },
2228 [IFLA_VXLAN_LINK] = { .type = NLA_U32 },
2229 [IFLA_VXLAN_LOCAL] = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2230 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) },
2231 [IFLA_VXLAN_TOS] = { .type = NLA_U8 },
2232 [IFLA_VXLAN_TTL] = { .type = NLA_U8 },
2233 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 },
2234 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 },
2235 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 },
2236 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) },
2237 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 },
2238 [IFLA_VXLAN_RSC] = { .type = NLA_U8 },
2239 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 },
2240 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 },
2241 [IFLA_VXLAN_PORT] = { .type = NLA_U16 },
2242 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 },
2243 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 },
2244 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 },
2247 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2249 if (tb[IFLA_ADDRESS]) {
2250 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2251 pr_debug("invalid link address (not ethernet)\n");
2255 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2256 pr_debug("invalid all zero ethernet address\n");
2257 return -EADDRNOTAVAIL;
2264 if (data[IFLA_VXLAN_ID]) {
2265 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2266 if (id >= VXLAN_VID_MASK)
2270 if (data[IFLA_VXLAN_PORT_RANGE]) {
2271 const struct ifla_vxlan_port_range *p
2272 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2274 if (ntohs(p->high) < ntohs(p->low)) {
2275 pr_debug("port range %u .. %u not valid\n",
2276 ntohs(p->low), ntohs(p->high));
2284 static void vxlan_get_drvinfo(struct net_device *netdev,
2285 struct ethtool_drvinfo *drvinfo)
2287 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2288 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2291 static const struct ethtool_ops vxlan_ethtool_ops = {
2292 .get_drvinfo = vxlan_get_drvinfo,
2293 .get_link = ethtool_op_get_link,
2296 static void vxlan_del_work(struct work_struct *work)
2298 struct vxlan_sock *vs = container_of(work, struct vxlan_sock, del_work);
2299 udp_tunnel_sock_release(vs->sock);
2303 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2304 __be16 port, u32 flags)
2306 struct socket *sock;
2307 struct udp_port_cfg udp_conf;
2310 memset(&udp_conf, 0, sizeof(udp_conf));
2313 udp_conf.family = AF_INET6;
2314 udp_conf.use_udp6_tx_checksums =
2315 !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2316 udp_conf.use_udp6_rx_checksums =
2317 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2319 udp_conf.family = AF_INET;
2320 udp_conf.local_ip.s_addr = INADDR_ANY;
2321 udp_conf.use_udp_checksums =
2322 !!(flags & VXLAN_F_UDP_CSUM);
2325 udp_conf.local_udp_port = port;
2327 /* Open UDP socket */
2328 err = udp_sock_create(net, &udp_conf, &sock);
2330 return ERR_PTR(err);
2335 /* Create new listen socket if needed */
2336 static struct vxlan_sock *vxlan_socket_create(struct net *net, __be16 port,
2337 vxlan_rcv_t *rcv, void *data,
2340 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2341 struct vxlan_sock *vs;
2342 struct socket *sock;
2344 bool ipv6 = !!(flags & VXLAN_F_IPV6);
2345 struct udp_tunnel_sock_cfg tunnel_cfg;
2347 vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2349 return ERR_PTR(-ENOMEM);
2351 for (h = 0; h < VNI_HASH_SIZE; ++h)
2352 INIT_HLIST_HEAD(&vs->vni_list[h]);
2354 INIT_WORK(&vs->del_work, vxlan_del_work);
2356 sock = vxlan_create_sock(net, ipv6, port, flags);
2359 return ERR_CAST(sock);
2363 atomic_set(&vs->refcnt, 1);
2367 /* Initialize the vxlan udp offloads structure */
2368 vs->udp_offloads.port = port;
2369 vs->udp_offloads.callbacks.gro_receive = vxlan_gro_receive;
2370 vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
2372 spin_lock(&vn->sock_lock);
2373 hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2374 vxlan_notify_add_rx_port(vs);
2375 spin_unlock(&vn->sock_lock);
2377 /* Mark socket as an encapsulation socket. */
2378 tunnel_cfg.sk_user_data = vs;
2379 tunnel_cfg.encap_type = 1;
2380 tunnel_cfg.encap_rcv = vxlan_udp_encap_recv;
2381 tunnel_cfg.encap_destroy = NULL;
2383 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2388 struct vxlan_sock *vxlan_sock_add(struct net *net, __be16 port,
2389 vxlan_rcv_t *rcv, void *data,
2390 bool no_share, u32 flags)
2392 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2393 struct vxlan_sock *vs;
2394 bool ipv6 = flags & VXLAN_F_IPV6;
2396 vs = vxlan_socket_create(net, port, rcv, data, flags);
2400 if (no_share) /* Return error if sharing is not allowed. */
2403 spin_lock(&vn->sock_lock);
2404 vs = vxlan_find_sock(net, ipv6 ? AF_INET6 : AF_INET, port);
2405 if (vs && ((vs->rcv != rcv) ||
2406 !atomic_add_unless(&vs->refcnt, 1, 0)))
2407 vs = ERR_PTR(-EBUSY);
2408 spin_unlock(&vn->sock_lock);
2411 vs = ERR_PTR(-EINVAL);
2415 EXPORT_SYMBOL_GPL(vxlan_sock_add);
2417 /* Scheduled at device creation to bind to a socket */
2418 static void vxlan_sock_work(struct work_struct *work)
2420 struct vxlan_dev *vxlan = container_of(work, struct vxlan_dev, sock_work);
2421 struct net *net = vxlan->net;
2422 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2423 __be16 port = vxlan->dst_port;
2424 struct vxlan_sock *nvs;
2426 nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags);
2427 spin_lock(&vn->sock_lock);
2429 vxlan_vs_add_dev(nvs, vxlan);
2430 spin_unlock(&vn->sock_lock);
2432 dev_put(vxlan->dev);
2435 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2436 struct nlattr *tb[], struct nlattr *data[])
2438 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2439 struct vxlan_dev *vxlan = netdev_priv(dev);
2440 struct vxlan_rdst *dst = &vxlan->default_dst;
2443 bool use_ipv6 = false;
2445 if (!data[IFLA_VXLAN_ID])
2448 vxlan->net = src_net;
2450 vni = nla_get_u32(data[IFLA_VXLAN_ID]);
2451 dst->remote_vni = vni;
2453 /* Unless IPv6 is explicitly requested, assume IPv4 */
2454 dst->remote_ip.sa.sa_family = AF_INET;
2455 if (data[IFLA_VXLAN_GROUP]) {
2456 dst->remote_ip.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_GROUP]);
2457 } else if (data[IFLA_VXLAN_GROUP6]) {
2458 if (!IS_ENABLED(CONFIG_IPV6))
2459 return -EPFNOSUPPORT;
2461 nla_memcpy(&dst->remote_ip.sin6.sin6_addr, data[IFLA_VXLAN_GROUP6],
2462 sizeof(struct in6_addr));
2463 dst->remote_ip.sa.sa_family = AF_INET6;
2467 if (data[IFLA_VXLAN_LOCAL]) {
2468 vxlan->saddr.sin.sin_addr.s_addr = nla_get_be32(data[IFLA_VXLAN_LOCAL]);
2469 vxlan->saddr.sa.sa_family = AF_INET;
2470 } else if (data[IFLA_VXLAN_LOCAL6]) {
2471 if (!IS_ENABLED(CONFIG_IPV6))
2472 return -EPFNOSUPPORT;
2474 /* TODO: respect scope id */
2475 nla_memcpy(&vxlan->saddr.sin6.sin6_addr, data[IFLA_VXLAN_LOCAL6],
2476 sizeof(struct in6_addr));
2477 vxlan->saddr.sa.sa_family = AF_INET6;
2481 if (data[IFLA_VXLAN_LINK] &&
2482 (dst->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]))) {
2483 struct net_device *lowerdev
2484 = __dev_get_by_index(src_net, dst->remote_ifindex);
2487 pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2491 #if IS_ENABLED(CONFIG_IPV6)
2493 struct inet6_dev *idev = __in6_dev_get(lowerdev);
2494 if (idev && idev->cnf.disable_ipv6) {
2495 pr_info("IPv6 is disabled via sysctl\n");
2498 vxlan->flags |= VXLAN_F_IPV6;
2503 dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2505 dev->needed_headroom = lowerdev->hard_header_len +
2506 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2507 } else if (use_ipv6)
2508 vxlan->flags |= VXLAN_F_IPV6;
2510 if (data[IFLA_VXLAN_TOS])
2511 vxlan->tos = nla_get_u8(data[IFLA_VXLAN_TOS]);
2513 if (data[IFLA_VXLAN_TTL])
2514 vxlan->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2516 if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2517 vxlan->flags |= VXLAN_F_LEARN;
2519 if (data[IFLA_VXLAN_AGEING])
2520 vxlan->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2522 vxlan->age_interval = FDB_AGE_DEFAULT;
2524 if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2525 vxlan->flags |= VXLAN_F_PROXY;
2527 if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2528 vxlan->flags |= VXLAN_F_RSC;
2530 if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2531 vxlan->flags |= VXLAN_F_L2MISS;
2533 if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2534 vxlan->flags |= VXLAN_F_L3MISS;
2536 if (data[IFLA_VXLAN_LIMIT])
2537 vxlan->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2539 if (data[IFLA_VXLAN_PORT_RANGE]) {
2540 const struct ifla_vxlan_port_range *p
2541 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2542 vxlan->port_min = ntohs(p->low);
2543 vxlan->port_max = ntohs(p->high);
2546 if (data[IFLA_VXLAN_PORT])
2547 vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2549 if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2550 vxlan->flags |= VXLAN_F_UDP_CSUM;
2552 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2553 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2554 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2556 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
2557 nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
2558 vxlan->flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
2560 if (vxlan_find_vni(src_net, vni, use_ipv6 ? AF_INET6 : AF_INET,
2562 pr_info("duplicate VNI %u\n", vni);
2566 dev->ethtool_ops = &vxlan_ethtool_ops;
2568 /* create an fdb entry for a valid default destination */
2569 if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2570 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2571 &vxlan->default_dst.remote_ip,
2572 NUD_REACHABLE|NUD_PERMANENT,
2573 NLM_F_EXCL|NLM_F_CREATE,
2575 vxlan->default_dst.remote_vni,
2576 vxlan->default_dst.remote_ifindex,
2582 err = register_netdevice(dev);
2584 vxlan_fdb_delete_default(vxlan);
2588 list_add(&vxlan->next, &vn->vxlan_list);
2593 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
2595 struct vxlan_dev *vxlan = netdev_priv(dev);
2596 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2598 spin_lock(&vn->sock_lock);
2599 if (!hlist_unhashed(&vxlan->hlist))
2600 hlist_del_rcu(&vxlan->hlist);
2601 spin_unlock(&vn->sock_lock);
2603 list_del(&vxlan->next);
2604 unregister_netdevice_queue(dev, head);
2607 static size_t vxlan_get_size(const struct net_device *dev)
2610 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */
2611 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
2612 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
2613 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
2614 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */
2615 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */
2616 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */
2617 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */
2618 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */
2619 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */
2620 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */
2621 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2622 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2623 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2624 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
2625 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
2626 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
2627 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2631 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
2633 const struct vxlan_dev *vxlan = netdev_priv(dev);
2634 const struct vxlan_rdst *dst = &vxlan->default_dst;
2635 struct ifla_vxlan_port_range ports = {
2636 .low = htons(vxlan->port_min),
2637 .high = htons(vxlan->port_max),
2640 if (nla_put_u32(skb, IFLA_VXLAN_ID, dst->remote_vni))
2641 goto nla_put_failure;
2643 if (!vxlan_addr_any(&dst->remote_ip)) {
2644 if (dst->remote_ip.sa.sa_family == AF_INET) {
2645 if (nla_put_be32(skb, IFLA_VXLAN_GROUP,
2646 dst->remote_ip.sin.sin_addr.s_addr))
2647 goto nla_put_failure;
2648 #if IS_ENABLED(CONFIG_IPV6)
2650 if (nla_put(skb, IFLA_VXLAN_GROUP6, sizeof(struct in6_addr),
2651 &dst->remote_ip.sin6.sin6_addr))
2652 goto nla_put_failure;
2657 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
2658 goto nla_put_failure;
2660 if (!vxlan_addr_any(&vxlan->saddr)) {
2661 if (vxlan->saddr.sa.sa_family == AF_INET) {
2662 if (nla_put_be32(skb, IFLA_VXLAN_LOCAL,
2663 vxlan->saddr.sin.sin_addr.s_addr))
2664 goto nla_put_failure;
2665 #if IS_ENABLED(CONFIG_IPV6)
2667 if (nla_put(skb, IFLA_VXLAN_LOCAL6, sizeof(struct in6_addr),
2668 &vxlan->saddr.sin6.sin6_addr))
2669 goto nla_put_failure;
2674 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->ttl) ||
2675 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->tos) ||
2676 nla_put_u8(skb, IFLA_VXLAN_LEARNING,
2677 !!(vxlan->flags & VXLAN_F_LEARN)) ||
2678 nla_put_u8(skb, IFLA_VXLAN_PROXY,
2679 !!(vxlan->flags & VXLAN_F_PROXY)) ||
2680 nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
2681 nla_put_u8(skb, IFLA_VXLAN_L2MISS,
2682 !!(vxlan->flags & VXLAN_F_L2MISS)) ||
2683 nla_put_u8(skb, IFLA_VXLAN_L3MISS,
2684 !!(vxlan->flags & VXLAN_F_L3MISS)) ||
2685 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) ||
2686 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) ||
2687 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port) ||
2688 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
2689 !!(vxlan->flags & VXLAN_F_UDP_CSUM)) ||
2690 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
2691 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
2692 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
2693 !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)))
2694 goto nla_put_failure;
2696 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
2697 goto nla_put_failure;
2705 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
2707 .maxtype = IFLA_VXLAN_MAX,
2708 .policy = vxlan_policy,
2709 .priv_size = sizeof(struct vxlan_dev),
2710 .setup = vxlan_setup,
2711 .validate = vxlan_validate,
2712 .newlink = vxlan_newlink,
2713 .dellink = vxlan_dellink,
2714 .get_size = vxlan_get_size,
2715 .fill_info = vxlan_fill_info,
2718 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
2719 struct net_device *dev)
2721 struct vxlan_dev *vxlan, *next;
2722 LIST_HEAD(list_kill);
2724 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2725 struct vxlan_rdst *dst = &vxlan->default_dst;
2727 /* In case we created vxlan device with carrier
2728 * and we loose the carrier due to module unload
2729 * we also need to remove vxlan device. In other
2730 * cases, it's not necessary and remote_ifindex
2731 * is 0 here, so no matches.
2733 if (dst->remote_ifindex == dev->ifindex)
2734 vxlan_dellink(vxlan->dev, &list_kill);
2737 unregister_netdevice_many(&list_kill);
2740 static int vxlan_lowerdev_event(struct notifier_block *unused,
2741 unsigned long event, void *ptr)
2743 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2744 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
2746 if (event == NETDEV_UNREGISTER)
2747 vxlan_handle_lowerdev_unregister(vn, dev);
2752 static struct notifier_block vxlan_notifier_block __read_mostly = {
2753 .notifier_call = vxlan_lowerdev_event,
2756 static __net_init int vxlan_init_net(struct net *net)
2758 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2761 INIT_LIST_HEAD(&vn->vxlan_list);
2762 spin_lock_init(&vn->sock_lock);
2764 for (h = 0; h < PORT_HASH_SIZE; ++h)
2765 INIT_HLIST_HEAD(&vn->sock_list[h]);
2770 static void __net_exit vxlan_exit_net(struct net *net)
2772 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2773 struct vxlan_dev *vxlan, *next;
2774 struct net_device *dev, *aux;
2778 for_each_netdev_safe(net, dev, aux)
2779 if (dev->rtnl_link_ops == &vxlan_link_ops)
2780 unregister_netdevice_queue(dev, &list);
2782 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
2783 /* If vxlan->dev is in the same netns, it has already been added
2784 * to the list by the previous loop.
2786 if (!net_eq(dev_net(vxlan->dev), net))
2787 unregister_netdevice_queue(dev, &list);
2790 unregister_netdevice_many(&list);
2794 static struct pernet_operations vxlan_net_ops = {
2795 .init = vxlan_init_net,
2796 .exit = vxlan_exit_net,
2797 .id = &vxlan_net_id,
2798 .size = sizeof(struct vxlan_net),
2801 static int __init vxlan_init_module(void)
2805 vxlan_wq = alloc_workqueue("vxlan", 0, 0);
2809 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
2811 rc = register_pernet_subsys(&vxlan_net_ops);
2815 rc = register_netdevice_notifier(&vxlan_notifier_block);
2819 rc = rtnl_link_register(&vxlan_link_ops);
2825 unregister_netdevice_notifier(&vxlan_notifier_block);
2827 unregister_pernet_subsys(&vxlan_net_ops);
2829 destroy_workqueue(vxlan_wq);
2832 late_initcall(vxlan_init_module);
2834 static void __exit vxlan_cleanup_module(void)
2836 rtnl_link_unregister(&vxlan_link_ops);
2837 unregister_netdevice_notifier(&vxlan_notifier_block);
2838 destroy_workqueue(vxlan_wq);
2839 unregister_pernet_subsys(&vxlan_net_ops);
2840 /* rcu_barrier() is called by netns */
2842 module_exit(vxlan_cleanup_module);
2844 MODULE_LICENSE("GPL");
2845 MODULE_VERSION(VXLAN_VERSION);
2846 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
2847 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
2848 MODULE_ALIAS_RTNL_LINK("vxlan");